September 23, 2026

Four new AI models and the job each one fits

New AI models, September 2026

How we scored and priced each model

Every score below comes from the Artificial Analysis Intelligence Index v4.3.2, which combines 10 independent evaluations. Cost is Artificial Analysis’s cost per index task: what it paid, on average, to get one task done. That figure includes the list price and how much the model writes to finish the job, so a cheap model that writes a lot can still cost more per task.

Each chart plots score against cost per task on a log scale. The shaded best value zone covers models that score 45 or higher for $1.50 or less per task. The line joins the models no one else beats on both score and cost.

Each model is shown at its highest scoring setting on Artificial Analysis: maximum reasoning for Claude Opus 5.5, GPT-6 Sol and GPT-6 Luna, and “xhigh” for Grok 4.7.

Claude Opus 5.5: for work that has to be right

Claude Opus 5.5 is the new top model on the index. It scores 58, five points ahead of GPT-6 Astra and Claude Fable 5.1, the next best models.

Opus 5.5 also costs $5.98 per index task, more than 5 times GPT-6 Sol’s $1.06. Its list price is $4 per million tokens in and $20 out, but it writes long, thorough answers, and that output drives up the cost of each task.

• Reads up to 1M tokens in a single request and writes up to 128K

• Released by Anthropic on September 22, 2026

• Runs on the Claude API, Amazon Bedrock, Google Vertex AI and Microsoft Foundry

Use it for complex analysis, large code changes and reviews where a mistake is expensive. Route routine volume to a cheaper model.

GPT-6 Sol: the everyday default

GPT-6 Sol scores 48 on the index for $1.06 per task, under a fifth of Claude Opus 5.5’s cost. It sits inside the best value zone and on the value frontier: no model on the chart scores higher for less.

It also improves on the model it replaces. GPT-5.6 Sol scored 47 at $1.99 per task, so GPT-6 Sol delivers a higher score at about half the cost.

• Released by OpenAI on September 22, 2026

• Reads up to 1.05M tokens in a single request and writes up to 128K

• Priced at $2 per million tokens in and $10 out, with higher rates for prompts over 272K tokens

• Writes output about 3 times faster than Grok 4.7, per Artificial Analysis

Opus 5.5 scores 10 points higher, so keep the hardest work there. For drafting, document summaries, ticket triage and most business automation, GPT-6 Sol gets the job done at a fraction of the cost.

GPT-6 Luna: for volume

GPT-6 Luna costs about 7 cents per task. It matches GPT-5.6 Luna’s score of 37 for 61% less per task ($0.07 against $0.18).

• Released by OpenAI on September 22, 2026, alongside GPT-6 Sol

• Priced at $0.10 per million tokens in and $0.50 out

• Same context limits as GPT-6 Sol: 1.05M tokens in, 128K out

• Writes output faster than GPT-6 Sol, per Artificial Analysis

A score of 37 puts it well below the flagships. Use it for sorting, tagging, extraction and routing, where you run millions of simple tasks and cost per task decides the budget.

Grok 4.7: SpaceXAI’s new coding flagship

Grok 4.7 is SpaceXAI’s new coding flagship, released September 21, 2026. It scores 46 on the index, up from 44 for Grok 4.6.

• Reads up to 500K tokens in a single request

• Available in Cursor, Grok Build and the Grok API

• Priced at $2 per million tokens in and $6 out, with higher rates for prompts of 200K tokens or more

Grok 4.7 writes long answers, so it costs $3.74 per index task, about twice Grok 4.6. On this index, GPT-6 Sol scores higher for less. Grok 4.7 fits developers who already work in Cursor or on SpaceXAI’s platform and want its newest model there.

Score and cost comparison: all four models

Model	Best for	Intelligence Index	Cost per task
Claude Opus 5.5	Hard, high-stakes work	58	$5.98
GPT-6 Sol	Everyday automation	48	$1.06
Grok 4.7	Coding in Cursor and the Grok API	46	$3.74
GPT-6 Luna	Simple tasks at volume	37	$0.07

Scores and costs: Artificial Analysis Intelligence Index v4.3.2, September 2026.

Run them on your own workflows

Claude Opus 5.5, GPT-6 Sol, GPT-6 Luna and Grok 4.7 are available for enterprise teams on Complete. Book a demo and we will run them on your own work.